The below answer was provided by Steve Byrnes, Postdoc in Physics (2014), Harvard University on Quora and the same answer has been featured in Forbes.
The second law of thermodynamics forbids a 100%-efficient solar cell. More specifically, Carnot’s theorem applies to photovoltaics and any other solar energy system, where the hot side of the “heat engine” is the temperature of the sun and the cold side is the ambient temperature on earth. (This is slightly oversimplified.)
The result is, for a system with sunlight concentration (lenses and mirrors and motors to follow the sun as it moves in the sky), the maximum efficiency is ~85%, and for a system that does not track the sun, the maximum efficiency is ~55%. (For details see my calculations here.)
On an overcast day, tracking the sun doesn’t work, so ~55% is the theoretical maximum.
On the market today, the highest efficiency that money can buy is … drumroll … ~35% for unconcentrated photovoltaics (PV) (e.g. Spectrolab), ~35% for concentrated PV (e.g. Amonix), and ~35% for solar thermal (e.g. Ripasso).
By the way, in unconcentrated PV, there is currently a huge gap between the highest efficiency that money can buy (~35% from Spectrolab, for ~$100,000 per square meter) and the highest efficiency that is not insanely expensive (~20% silicon modules from SunPower SPWR -4.35%).
I expect that gap to shrink dramatically in the next NXGPY +% 10-20 years thanks to Alta Devices, which already has a pilot line creating affordable ~25%-efficient solar modules, and is moving towards 30% or even beyond. These cells will be light and flexible too! This is very exciting. But I’m getting off-topic. The question is not primarily about what’s affordable, but what’s possible. How to explain the gap between ~35% and the theoretical maximum?
For unconcentrated PV, the best cells (currently ~35%) have been creeping towards the theoretical maximum (~55%) for decades (see chart), and I expect they will continue to do so. I don’t mean that they will literally asymptotically approach closer and closer to 55%; eventually there will be a tradeoff where higher nominal efficiency (under standard test conditions) comes at the expense of lower real-world efficiency (which involves working robustly under a variety of light and temperature conditions).
So there is a ceiling for unconcentrated PV efficiency, and it’s somewhere between ~35% and ~55%, but I don’t know where.
For concentrated PV: In theory, PV cells should get more and more efficient as light concentration increases. In other words, if you double the light intensity, it should *more* than double the electricity generation. That’s why the theoretical limit for concentrated systems (~85%) is higher than unconcentrated (~55%). However, there is a cost to concentration too:
(1) The lenses / mirrors are not perfect;
(2) The solar cell will get hotter, which lowers its efficiency;
(3) You can only get power out of the light coming directly from the sun, not the diffuse blue light from the rest of the sky, which accounts for at least 15% of the light, sometimes more.
Thanks to those problems, the best concentrated PV system that money can buy is more-or-less equally efficient as the best unconcentrated system that money can buy. Will that always be true? Well, the nominal theoretical limit is ~85%, but the only way to get that high is to concentrate sunlight to the maximum possible concentration of 50,000X.
At a more realistic concentration like 1000X, the theoretical limit is ~75%. Next, we account for the 15% or more diffuse light, and we’re down to ~65%. After accounting for imperfect lenses/mirrors and cell heating, we are probably down to a limit of 55-60%. So, I don’t think we should expect a huge divergence between the best available concentrated PV versus unconcentrated PV. The efficiency will be basically determined by the PV cell, and the concentrator will have only a small effect on the system-level efficiency.
The final main category is solar thermal, which uses lenses and mirrors and solar-tracking to heat something really hot, and then use that to run a heat engine. The highest-efficiency solar thermal systems available today are based on stirling engines and are ~35% efficient.
A Stirling engine can already run near the Carnot limit, so presumably the primary way to increase efficiency of a solar thermal system is to heat the thing to a higher temperature. To get that theoretical ~85% efficiency, you need to concentrate the sunlight by a factor of 50,000, and heat the thing to 2000C. This temperature is insanely high: I think that no one knows how to make a long-lasting high-efficiency heat engine that can work at such a high temperature.
If you heat to “only” 1000C, the maximum efficiency drops to ~75%; if you heat to 600C — which is realistic in a solar stirling engine system — then the maximum efficiency is ~65% (or ~55% including the wasted 15% diffuse light, as discussed above).
That 55% figure is still way above the ~35% that has been achieved to date, so there seems to be plenty of room for improvement if the solar thermal industry continues to grow. But the 85% figure will never happen, and even 70% is extremely unlikely.
(For completeness, I should mention that there are solar power systems that don’t fit in any of the above categories, like thermophotonics and thermophotovoltaics. These are very early-stage ideas, and I don’t know enough about them to comment.)